K2 Horizon 375B-A23B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
The flagship of the K2 Horizon family: a sparse MoE that stores 375B parameters and runs 23B per token (192 experts, 8 routed), with a native 524,288-token context. Apache-2.0, released with its data, training code and intermediate checkpoints. The checkpoint holds 379.2B tensors, which is what must be resident. The model card's serving routes are vLLM and SGLang, validated on 8× H200.
K2 Horizon 375B-A23B needs about 230 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 127.5 GB | — | Too big | |
| Q3_K_M 3.41 bpw | 164.5 GB | — | Too big | |
| Q4_K_M 4.83 bpw | 231.8 GB | — | Too big | |
| Q5_K_M 5.67 bpw | 271.6 GB | — | Too big | |
| Q6_K 6.56 bpw | 313.8 GB | — | Too big | |
| Q8_0 8.50 bpw | 405.7 GB | — | Too big | |
| F16 16.00 bpw | 761.2 GB | — | Too big |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. K2 Horizon 375B-A23B VRAM calculator →
Get K2 Horizon 375B-A23B running
The cheapest catalogued GPU that runs K2 Horizon 375B-A23B is the Apple M3 Ultra (512 GB).
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run K2 Horizon 375B-A23B
No Ollama library tag and no published GGUF for this size. The model card's serving routes are vLLM and SGLang.
Specifications
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- Parameters
- 375 Billion (23B active)
- Context window
- 524,288
- Architecture
- Sparse MoE (192 experts, 8 active), GQA
- Provider
- MBZUAI Institute of Foundation Models
- Licence
- Apache 2.0
- Availability
- Weights downloadable
- Practical deployment
- Workstation (large unified memory or pro GPU)
- Specified at
- Q4_K_M
- System RAM
- 256 GB
- Record updated
- 2026-10-13
- Runs with
- vLLM: supported SGLang: supported Ollama: not supported
Commercial use permitted. No usage restrictions beyond attribution.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Independent evaluation — not vendor-claimed and not measured by us
| Benchmark | Score | Provenance |
|---|---|---|
| AA Intelligence Index v4.3.2 | 31 | independent · Artificial Analysis · snapshot 2026-10-13 |
Other K2 Horizon sizes
K2 Horizon 375B-A23B — frequently asked questions
How much VRAM does K2 Horizon 375B-A23B need?
About 230 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does K2 Horizon 375B-A23B run on an RTX 4090 (24 GB)?
No. K2 Horizon 375B-A23B needs about 230 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.
How do I run K2 Horizon 375B-A23B locally?
No Ollama library tag and no published GGUF for this size. The model card's serving routes are vLLM and SGLang. Running the published tag would send your prompts to a hosted GPU rather than your own machine.
What other sizes does K2 Horizon come in?
K2 Horizon 3.7B (4 GB), K2 Horizon 7B (6 GB), K2 Horizon MoVA 36B-A4B (23 GB), K2 Horizon 375B-A23B (230 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.